?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: rockchip: fix broken build
[linux/fpc-iii.git] / fs / nfs / pnfs.c
blob70bf706b10904e156affe9dd4bea2ec9a17776c5
1 /*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_page.h>
32 #include <linux/module.h>
33 #include "internal.h"
34 #include "pnfs.h"
35 #include "iostat.h"
36 #include "nfs4trace.h"
37 #include "delegation.h"
38 #include "nfs42.h"
39
40 #define NFSDBG_FACILITY NFSDBG_PNFS
41 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
42
43 /* Locking:
44 *
45 * pnfs_spinlock:
46 * protects pnfs_modules_tbl.
47 */
48 static DEFINE_SPINLOCK(pnfs_spinlock);
49
50 /*
51 * pnfs_modules_tbl holds all pnfs modules
52 */
53 static LIST_HEAD(pnfs_modules_tbl);
54
55 static int
56 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
57 enum pnfs_iomode iomode, bool sync);
58
59 /* Return the registered pnfs layout driver module matching given id */
60 static struct pnfs_layoutdriver_type *
61 find_pnfs_driver_locked(u32 id)
62 {
63 struct pnfs_layoutdriver_type *local;
64
65 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
66 if (local->id == id)
67 goto out;
68 local = NULL;
69 out:
70 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
71 return local;
72 }
73
74 static struct pnfs_layoutdriver_type *
75 find_pnfs_driver(u32 id)
76 {
77 struct pnfs_layoutdriver_type *local;
78
79 spin_lock(&pnfs_spinlock);
80 local = find_pnfs_driver_locked(id);
81 if (local != NULL && !try_module_get(local->owner)) {
82 dprintk("%s: Could not grab reference on module\n", __func__);
83 local = NULL;
84 }
85 spin_unlock(&pnfs_spinlock);
86 return local;
87 }
88
89 void
90 unset_pnfs_layoutdriver(struct nfs_server *nfss)
91 {
92 if (nfss->pnfs_curr_ld) {
93 if (nfss->pnfs_curr_ld->clear_layoutdriver)
94 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
95 /* Decrement the MDS count. Purge the deviceid cache if zero */
96 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
97 nfs4_deviceid_purge_client(nfss->nfs_client);
98 module_put(nfss->pnfs_curr_ld->owner);
99 }
100 nfss->pnfs_curr_ld = NULL;
101 }
102
103 /*
104 * Try to set the server's pnfs module to the pnfs layout type specified by id.
105 * Currently only one pNFS layout driver per filesystem is supported.
106 *
107 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
108 */
109 void
110 set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
111 u32 id)
112 {
113 struct pnfs_layoutdriver_type *ld_type = NULL;
114
115 if (id == 0)
116 goto out_no_driver;
117 if (!(server->nfs_client->cl_exchange_flags &
118 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
119 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
120 __func__, id, server->nfs_client->cl_exchange_flags);
121 goto out_no_driver;
122 }
123 ld_type = find_pnfs_driver(id);
124 if (!ld_type) {
125 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
126 ld_type = find_pnfs_driver(id);
127 if (!ld_type) {
128 dprintk("%s: No pNFS module found for %u.\n",
129 __func__, id);
130 goto out_no_driver;
131 }
132 }
133 server->pnfs_curr_ld = ld_type;
134 if (ld_type->set_layoutdriver
135 && ld_type->set_layoutdriver(server, mntfh)) {
136 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
137 "driver %u.\n", __func__, id);
138 module_put(ld_type->owner);
139 goto out_no_driver;
140 }
141 /* Bump the MDS count */
142 atomic_inc(&server->nfs_client->cl_mds_count);
143
144 dprintk("%s: pNFS module for %u set\n", __func__, id);
145 return;
146
147 out_no_driver:
148 dprintk("%s: Using NFSv4 I/O\n", __func__);
149 server->pnfs_curr_ld = NULL;
150 }
151
152 int
153 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
154 {
155 int status = -EINVAL;
156 struct pnfs_layoutdriver_type *tmp;
157
158 if (ld_type->id == 0) {
159 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
160 return status;
161 }
162 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
163 printk(KERN_ERR "NFS: %s Layout driver must provide "
164 "alloc_lseg and free_lseg.\n", __func__);
165 return status;
166 }
167
168 spin_lock(&pnfs_spinlock);
169 tmp = find_pnfs_driver_locked(ld_type->id);
170 if (!tmp) {
171 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
172 status = 0;
173 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
174 ld_type->name);
175 } else {
176 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
177 __func__, ld_type->id);
178 }
179 spin_unlock(&pnfs_spinlock);
180
181 return status;
182 }
183 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
184
185 void
186 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
187 {
188 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
189 spin_lock(&pnfs_spinlock);
190 list_del(&ld_type->pnfs_tblid);
191 spin_unlock(&pnfs_spinlock);
192 }
193 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
194
195 /*
196 * pNFS client layout cache
197 */
198
199 /* Need to hold i_lock if caller does not already hold reference */
200 void
201 pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
202 {
203 atomic_inc(&lo->plh_refcount);
204 }
205
206 static struct pnfs_layout_hdr *
207 pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
208 {
209 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
210 return ld->alloc_layout_hdr(ino, gfp_flags);
211 }
212
213 static void
214 pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
215 {
216 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
217 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
218
219 if (!list_empty(&lo->plh_layouts)) {
220 struct nfs_client *clp = server->nfs_client;
221
222 spin_lock(&clp->cl_lock);
223 list_del_init(&lo->plh_layouts);
224 spin_unlock(&clp->cl_lock);
225 }
226 put_rpccred(lo->plh_lc_cred);
227 return ld->free_layout_hdr(lo);
228 }
229
230 static void
231 pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
232 {
233 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
234 dprintk("%s: freeing layout cache %p\n", __func__, lo);
235 nfsi->layout = NULL;
236 /* Reset MDS Threshold I/O counters */
237 nfsi->write_io = 0;
238 nfsi->read_io = 0;
239 }
240
241 void
242 pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
243 {
244 struct inode *inode = lo->plh_inode;
245
246 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
247 if (!list_empty(&lo->plh_segs))
248 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
249 pnfs_detach_layout_hdr(lo);
250 spin_unlock(&inode->i_lock);
251 pnfs_free_layout_hdr(lo);
252 }
253 }
254
255 static int
256 pnfs_iomode_to_fail_bit(u32 iomode)
257 {
258 return iomode == IOMODE_RW ?
259 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
260 }
261
262 static void
263 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
264 {
265 lo->plh_retry_timestamp = jiffies;
266 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
267 atomic_inc(&lo->plh_refcount);
268 }
269
270 static void
271 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
272 {
273 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
274 atomic_dec(&lo->plh_refcount);
275 }
276
277 static void
278 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
279 {
280 struct inode *inode = lo->plh_inode;
281 struct pnfs_layout_range range = {
282 .iomode = iomode,
283 .offset = 0,
284 .length = NFS4_MAX_UINT64,
285 };
286 LIST_HEAD(head);
287
288 spin_lock(&inode->i_lock);
289 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
290 pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
291 spin_unlock(&inode->i_lock);
292 pnfs_free_lseg_list(&head);
293 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
294 iomode == IOMODE_RW ? "RW" : "READ");
295 }
296
297 static bool
298 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
299 {
300 unsigned long start, end;
301 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
302
303 if (test_bit(fail_bit, &lo->plh_flags) == 0)
304 return false;
305 end = jiffies;
306 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
307 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
308 /* It is time to retry the failed layoutgets */
309 pnfs_layout_clear_fail_bit(lo, fail_bit);
310 return false;
311 }
312 return true;
313 }
314
315 static void
316 init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
317 {
318 INIT_LIST_HEAD(&lseg->pls_list);
319 INIT_LIST_HEAD(&lseg->pls_lc_list);
320 atomic_set(&lseg->pls_refcount, 1);
321 smp_mb();
322 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
323 lseg->pls_layout = lo;
324 }
325
326 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
327 {
328 struct inode *ino = lseg->pls_layout->plh_inode;
329
330 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
331 }
332
333 static void
334 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
335 struct pnfs_layout_segment *lseg)
336 {
337 struct inode *inode = lo->plh_inode;
338
339 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
340 list_del_init(&lseg->pls_list);
341 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
342 atomic_dec(&lo->plh_refcount);
343 if (list_empty(&lo->plh_segs))
344 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
345 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
346 }
347
348 /* Return true if layoutreturn is needed */
349 static bool
350 pnfs_layout_need_return(struct pnfs_layout_hdr *lo,
351 struct pnfs_layout_segment *lseg)
352 {
353 struct pnfs_layout_segment *s;
354
355 if (!test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
356 return false;
357
358 list_for_each_entry(s, &lo->plh_segs, pls_list)
359 if (s != lseg && test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
360 return false;
361
362 return true;
363 }
364
365 static bool
366 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
367 {
368 if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
369 return false;
370 lo->plh_return_iomode = 0;
371 lo->plh_block_lgets++;
372 pnfs_get_layout_hdr(lo);
373 clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
374 return true;
375 }
376
377 static void pnfs_layoutreturn_before_put_lseg(struct pnfs_layout_segment *lseg,
378 struct pnfs_layout_hdr *lo, struct inode *inode)
379 {
380 lo = lseg->pls_layout;
381 inode = lo->plh_inode;
382
383 spin_lock(&inode->i_lock);
384 if (pnfs_layout_need_return(lo, lseg)) {
385 nfs4_stateid stateid;
386 enum pnfs_iomode iomode;
387 bool send;
388
389 stateid = lo->plh_stateid;
390 iomode = lo->plh_return_iomode;
391 send = pnfs_prepare_layoutreturn(lo);
392 spin_unlock(&inode->i_lock);
393 if (send) {
394 /* Send an async layoutreturn so we dont deadlock */
395 pnfs_send_layoutreturn(lo, stateid, iomode, false);
396 }
397 } else
398 spin_unlock(&inode->i_lock);
399 }
400
401 void
402 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
403 {
404 struct pnfs_layout_hdr *lo;
405 struct inode *inode;
406
407 if (!lseg)
408 return;
409
410 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
411 atomic_read(&lseg->pls_refcount),
412 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
413
414 /* Handle the case where refcount != 1 */
415 if (atomic_add_unless(&lseg->pls_refcount, -1, 1))
416 return;
417
418 lo = lseg->pls_layout;
419 inode = lo->plh_inode;
420 /* Do we need a layoutreturn? */
421 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
422 pnfs_layoutreturn_before_put_lseg(lseg, lo, inode);
423
424 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
425 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
426 spin_unlock(&inode->i_lock);
427 return;
428 }
429 pnfs_get_layout_hdr(lo);
430 pnfs_layout_remove_lseg(lo, lseg);
431 spin_unlock(&inode->i_lock);
432 pnfs_free_lseg(lseg);
433 pnfs_put_layout_hdr(lo);
434 }
435 }
436 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
437
438 static void pnfs_free_lseg_async_work(struct work_struct *work)
439 {
440 struct pnfs_layout_segment *lseg;
441 struct pnfs_layout_hdr *lo;
442
443 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
444 lo = lseg->pls_layout;
445
446 pnfs_free_lseg(lseg);
447 pnfs_put_layout_hdr(lo);
448 }
449
450 static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
451 {
452 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
453 schedule_work(&lseg->pls_work);
454 }
455
456 void
457 pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
458 {
459 if (!lseg)
460 return;
461
462 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
463
464 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
465 atomic_read(&lseg->pls_refcount),
466 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
467 if (atomic_dec_and_test(&lseg->pls_refcount)) {
468 struct pnfs_layout_hdr *lo = lseg->pls_layout;
469 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
470 return;
471 pnfs_get_layout_hdr(lo);
472 pnfs_layout_remove_lseg(lo, lseg);
473 pnfs_free_lseg_async(lseg);
474 }
475 }
476 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
477
478 static u64
479 end_offset(u64 start, u64 len)
480 {
481 u64 end;
482
483 end = start + len;
484 return end >= start ? end : NFS4_MAX_UINT64;
485 }
486
487 /*
488 * is l2 fully contained in l1?
489 * start1 end1
490 * [----------------------------------)
491 * start2 end2
492 * [----------------)
493 */
494 static bool
495 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
496 const struct pnfs_layout_range *l2)
497 {
498 u64 start1 = l1->offset;
499 u64 end1 = end_offset(start1, l1->length);
500 u64 start2 = l2->offset;
501 u64 end2 = end_offset(start2, l2->length);
502
503 return (start1 <= start2) && (end1 >= end2);
504 }
505
506 /*
507 * is l1 and l2 intersecting?
508 * start1 end1
509 * [----------------------------------)
510 * start2 end2
511 * [----------------)
512 */
513 static bool
514 pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
515 const struct pnfs_layout_range *l2)
516 {
517 u64 start1 = l1->offset;
518 u64 end1 = end_offset(start1, l1->length);
519 u64 start2 = l2->offset;
520 u64 end2 = end_offset(start2, l2->length);
521
522 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
523 (end2 == NFS4_MAX_UINT64 || end2 > start1);
524 }
525
526 static bool
527 should_free_lseg(const struct pnfs_layout_range *lseg_range,
528 const struct pnfs_layout_range *recall_range)
529 {
530 return (recall_range->iomode == IOMODE_ANY ||
531 lseg_range->iomode == recall_range->iomode) &&
532 pnfs_lseg_range_intersecting(lseg_range, recall_range);
533 }
534
535 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
536 struct list_head *tmp_list)
537 {
538 if (!atomic_dec_and_test(&lseg->pls_refcount))
539 return false;
540 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
541 list_add(&lseg->pls_list, tmp_list);
542 return true;
543 }
544
545 /* Returns 1 if lseg is removed from list, 0 otherwise */
546 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
547 struct list_head *tmp_list)
548 {
549 int rv = 0;
550
551 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
552 /* Remove the reference keeping the lseg in the
553 * list. It will now be removed when all
554 * outstanding io is finished.
555 */
556 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
557 atomic_read(&lseg->pls_refcount));
558 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
559 rv = 1;
560 }
561 return rv;
562 }
563
564 /* Returns count of number of matching invalid lsegs remaining in list
565 * after call.
566 */
567 int
568 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
569 struct list_head *tmp_list,
570 struct pnfs_layout_range *recall_range)
571 {
572 struct pnfs_layout_segment *lseg, *next;
573 int invalid = 0, removed = 0;
574
575 dprintk("%s:Begin lo %p\n", __func__, lo);
576
577 if (list_empty(&lo->plh_segs))
578 return 0;
579 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
580 if (!recall_range ||
581 should_free_lseg(&lseg->pls_range, recall_range)) {
582 dprintk("%s: freeing lseg %p iomode %d "
583 "offset %llu length %llu\n", __func__,
584 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
585 lseg->pls_range.length);
586 invalid++;
587 removed += mark_lseg_invalid(lseg, tmp_list);
588 }
589 dprintk("%s:Return %i\n", __func__, invalid - removed);
590 return invalid - removed;
591 }
592
593 /* note free_me must contain lsegs from a single layout_hdr */
594 void
595 pnfs_free_lseg_list(struct list_head *free_me)
596 {
597 struct pnfs_layout_segment *lseg, *tmp;
598
599 if (list_empty(free_me))
600 return;
601
602 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
603 list_del(&lseg->pls_list);
604 pnfs_free_lseg(lseg);
605 }
606 }
607
608 void
609 pnfs_destroy_layout(struct nfs_inode *nfsi)
610 {
611 struct pnfs_layout_hdr *lo;
612 LIST_HEAD(tmp_list);
613
614 spin_lock(&nfsi->vfs_inode.i_lock);
615 lo = nfsi->layout;
616 if (lo) {
617 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
618 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
619 pnfs_get_layout_hdr(lo);
620 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
621 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
622 pnfs_clear_retry_layoutget(lo);
623 spin_unlock(&nfsi->vfs_inode.i_lock);
624 pnfs_free_lseg_list(&tmp_list);
625 pnfs_put_layout_hdr(lo);
626 } else
627 spin_unlock(&nfsi->vfs_inode.i_lock);
628 }
629 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
630
631 static bool
632 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
633 struct list_head *layout_list)
634 {
635 struct pnfs_layout_hdr *lo;
636 bool ret = false;
637
638 spin_lock(&inode->i_lock);
639 lo = NFS_I(inode)->layout;
640 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
641 pnfs_get_layout_hdr(lo);
642 list_add(&lo->plh_bulk_destroy, layout_list);
643 ret = true;
644 }
645 spin_unlock(&inode->i_lock);
646 return ret;
647 }
648
649 /* Caller must hold rcu_read_lock and clp->cl_lock */
650 static int
651 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
652 struct nfs_server *server,
653 struct list_head *layout_list)
654 {
655 struct pnfs_layout_hdr *lo, *next;
656 struct inode *inode;
657
658 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
659 inode = igrab(lo->plh_inode);
660 if (inode == NULL)
661 continue;
662 list_del_init(&lo->plh_layouts);
663 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
664 continue;
665 rcu_read_unlock();
666 spin_unlock(&clp->cl_lock);
667 iput(inode);
668 spin_lock(&clp->cl_lock);
669 rcu_read_lock();
670 return -EAGAIN;
671 }
672 return 0;
673 }
674
675 static int
676 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
677 bool is_bulk_recall)
678 {
679 struct pnfs_layout_hdr *lo;
680 struct inode *inode;
681 struct pnfs_layout_range range = {
682 .iomode = IOMODE_ANY,
683 .offset = 0,
684 .length = NFS4_MAX_UINT64,
685 };
686 LIST_HEAD(lseg_list);
687 int ret = 0;
688
689 while (!list_empty(layout_list)) {
690 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
691 plh_bulk_destroy);
692 dprintk("%s freeing layout for inode %lu\n", __func__,
693 lo->plh_inode->i_ino);
694 inode = lo->plh_inode;
695
696 pnfs_layoutcommit_inode(inode, false);
697
698 spin_lock(&inode->i_lock);
699 list_del_init(&lo->plh_bulk_destroy);
700 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
701 if (is_bulk_recall)
702 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
703 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
704 ret = -EAGAIN;
705 spin_unlock(&inode->i_lock);
706 pnfs_free_lseg_list(&lseg_list);
707 pnfs_put_layout_hdr(lo);
708 iput(inode);
709 }
710 return ret;
711 }
712
713 int
714 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
715 struct nfs_fsid *fsid,
716 bool is_recall)
717 {
718 struct nfs_server *server;
719 LIST_HEAD(layout_list);
720
721 spin_lock(&clp->cl_lock);
722 rcu_read_lock();
723 restart:
724 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
725 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
726 continue;
727 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
728 server,
729 &layout_list) != 0)
730 goto restart;
731 }
732 rcu_read_unlock();
733 spin_unlock(&clp->cl_lock);
734
735 if (list_empty(&layout_list))
736 return 0;
737 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
738 }
739
740 int
741 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
742 bool is_recall)
743 {
744 struct nfs_server *server;
745 LIST_HEAD(layout_list);
746
747 spin_lock(&clp->cl_lock);
748 rcu_read_lock();
749 restart:
750 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
751 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
752 server,
753 &layout_list) != 0)
754 goto restart;
755 }
756 rcu_read_unlock();
757 spin_unlock(&clp->cl_lock);
758
759 if (list_empty(&layout_list))
760 return 0;
761 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
762 }
763
764 /*
765 * Called by the state manger to remove all layouts established under an
766 * expired lease.
767 */
768 void
769 pnfs_destroy_all_layouts(struct nfs_client *clp)
770 {
771 nfs4_deviceid_mark_client_invalid(clp);
772 nfs4_deviceid_purge_client(clp);
773
774 pnfs_destroy_layouts_byclid(clp, false);
775 }
776
777 /*
778 * Compare 2 layout stateid sequence ids, to see which is newer,
779 * taking into account wraparound issues.
780 */
781 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
782 {
783 return (s32)(s1 - s2) > 0;
784 }
785
786 /* update lo->plh_stateid with new if is more recent */
787 void
788 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
789 bool update_barrier)
790 {
791 u32 oldseq, newseq, new_barrier;
792 int empty = list_empty(&lo->plh_segs);
793
794 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
795 newseq = be32_to_cpu(new->seqid);
796 if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
797 nfs4_stateid_copy(&lo->plh_stateid, new);
798 if (update_barrier) {
799 new_barrier = be32_to_cpu(new->seqid);
800 } else {
801 /* Because of wraparound, we want to keep the barrier
802 * "close" to the current seqids.
803 */
804 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
805 }
806 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
807 lo->plh_barrier = new_barrier;
808 }
809 }
810
811 static bool
812 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
813 const nfs4_stateid *stateid)
814 {
815 u32 seqid = be32_to_cpu(stateid->seqid);
816
817 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
818 }
819
820 static bool
821 pnfs_layout_returning(const struct pnfs_layout_hdr *lo,
822 struct pnfs_layout_range *range)
823 {
824 return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
825 (lo->plh_return_iomode == IOMODE_ANY ||
826 lo->plh_return_iomode == range->iomode);
827 }
828
829 /* lget is set to 1 if called from inside send_layoutget call chain */
830 static bool
831 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo,
832 struct pnfs_layout_range *range, int lget)
833 {
834 return lo->plh_block_lgets ||
835 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
836 (list_empty(&lo->plh_segs) &&
837 (atomic_read(&lo->plh_outstanding) > lget)) ||
838 pnfs_layout_returning(lo, range);
839 }
840
841 int
842 pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
843 struct pnfs_layout_range *range,
844 struct nfs4_state *open_state)
845 {
846 int status = 0;
847
848 dprintk("--> %s\n", __func__);
849 spin_lock(&lo->plh_inode->i_lock);
850 if (pnfs_layoutgets_blocked(lo, range, 1)) {
851 status = -EAGAIN;
852 } else if (!nfs4_valid_open_stateid(open_state)) {
853 status = -EBADF;
854 } else if (list_empty(&lo->plh_segs) ||
855 test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
856 int seq;
857
858 do {
859 seq = read_seqbegin(&open_state->seqlock);
860 nfs4_stateid_copy(dst, &open_state->stateid);
861 } while (read_seqretry(&open_state->seqlock, seq));
862 } else
863 nfs4_stateid_copy(dst, &lo->plh_stateid);
864 spin_unlock(&lo->plh_inode->i_lock);
865 dprintk("<-- %s\n", __func__);
866 return status;
867 }
868
869 /*
870 * Get layout from server.
871 * for now, assume that whole file layouts are requested.
872 * arg->offset: 0
873 * arg->length: all ones
874 */
875 static struct pnfs_layout_segment *
876 send_layoutget(struct pnfs_layout_hdr *lo,
877 struct nfs_open_context *ctx,
878 struct pnfs_layout_range *range,
879 gfp_t gfp_flags)
880 {
881 struct inode *ino = lo->plh_inode;
882 struct nfs_server *server = NFS_SERVER(ino);
883 struct nfs4_layoutget *lgp;
884 struct pnfs_layout_segment *lseg;
885
886 dprintk("--> %s\n", __func__);
887
888 lgp = kzalloc(sizeof(*lgp), gfp_flags);
889 if (lgp == NULL)
890 return NULL;
891
892 lgp->args.minlength = PAGE_CACHE_SIZE;
893 if (lgp->args.minlength > range->length)
894 lgp->args.minlength = range->length;
895 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
896 lgp->args.range = *range;
897 lgp->args.type = server->pnfs_curr_ld->id;
898 lgp->args.inode = ino;
899 lgp->args.ctx = get_nfs_open_context(ctx);
900 lgp->gfp_flags = gfp_flags;
901 lgp->cred = lo->plh_lc_cred;
902
903 /* Synchronously retrieve layout information from server and
904 * store in lseg.
905 */
906 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
907 if (IS_ERR(lseg)) {
908 switch (PTR_ERR(lseg)) {
909 case -ENOMEM:
910 case -ERESTARTSYS:
911 break;
912 default:
913 /* remember that LAYOUTGET failed and suspend trying */
914 pnfs_layout_io_set_failed(lo, range->iomode);
915 }
916 return NULL;
917 } else
918 pnfs_layout_clear_fail_bit(lo,
919 pnfs_iomode_to_fail_bit(range->iomode));
920
921 return lseg;
922 }
923
924 static void pnfs_clear_layoutcommit(struct inode *inode,
925 struct list_head *head)
926 {
927 struct nfs_inode *nfsi = NFS_I(inode);
928 struct pnfs_layout_segment *lseg, *tmp;
929
930 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
931 return;
932 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
933 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
934 continue;
935 pnfs_lseg_dec_and_remove_zero(lseg, head);
936 }
937 }
938
939 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
940 {
941 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
942 smp_mb__after_atomic();
943 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
944 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
945 }
946
947 static int
948 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
949 enum pnfs_iomode iomode, bool sync)
950 {
951 struct inode *ino = lo->plh_inode;
952 struct nfs4_layoutreturn *lrp;
953 int status = 0;
954
955 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
956 if (unlikely(lrp == NULL)) {
957 status = -ENOMEM;
958 spin_lock(&ino->i_lock);
959 lo->plh_block_lgets--;
960 pnfs_clear_layoutreturn_waitbit(lo);
961 rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq);
962 spin_unlock(&ino->i_lock);
963 pnfs_put_layout_hdr(lo);
964 goto out;
965 }
966
967 lrp->args.stateid = stateid;
968 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
969 lrp->args.inode = ino;
970 lrp->args.range.iomode = iomode;
971 lrp->args.range.offset = 0;
972 lrp->args.range.length = NFS4_MAX_UINT64;
973 lrp->args.layout = lo;
974 lrp->clp = NFS_SERVER(ino)->nfs_client;
975 lrp->cred = lo->plh_lc_cred;
976
977 status = nfs4_proc_layoutreturn(lrp, sync);
978 out:
979 dprintk("<-- %s status: %d\n", __func__, status);
980 return status;
981 }
982
983 /*
984 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
985 * when the layout segment list is empty.
986 *
987 * Note that a pnfs_layout_hdr can exist with an empty layout segment
988 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
989 * deviceid is marked invalid.
990 */
991 int
992 _pnfs_return_layout(struct inode *ino)
993 {
994 struct pnfs_layout_hdr *lo = NULL;
995 struct nfs_inode *nfsi = NFS_I(ino);
996 LIST_HEAD(tmp_list);
997 nfs4_stateid stateid;
998 int status = 0, empty;
999 bool send;
1000
1001 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1002
1003 spin_lock(&ino->i_lock);
1004 lo = nfsi->layout;
1005 if (!lo) {
1006 spin_unlock(&ino->i_lock);
1007 dprintk("NFS: %s no layout to return\n", __func__);
1008 goto out;
1009 }
1010 stateid = nfsi->layout->plh_stateid;
1011 /* Reference matched in nfs4_layoutreturn_release */
1012 pnfs_get_layout_hdr(lo);
1013 empty = list_empty(&lo->plh_segs);
1014 pnfs_clear_layoutcommit(ino, &tmp_list);
1015 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
1016
1017 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1018 struct pnfs_layout_range range = {
1019 .iomode = IOMODE_ANY,
1020 .offset = 0,
1021 .length = NFS4_MAX_UINT64,
1022 };
1023 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1024 }
1025
1026 /* Don't send a LAYOUTRETURN if list was initially empty */
1027 if (empty) {
1028 spin_unlock(&ino->i_lock);
1029 dprintk("NFS: %s no layout segments to return\n", __func__);
1030 goto out_put_layout_hdr;
1031 }
1032
1033 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1034 send = pnfs_prepare_layoutreturn(lo);
1035 spin_unlock(&ino->i_lock);
1036 pnfs_free_lseg_list(&tmp_list);
1037 if (send)
1038 status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
1039 out_put_layout_hdr:
1040 pnfs_put_layout_hdr(lo);
1041 out:
1042 dprintk("<-- %s status: %d\n", __func__, status);
1043 return status;
1044 }
1045 EXPORT_SYMBOL_GPL(_pnfs_return_layout);
1046
1047 int
1048 pnfs_commit_and_return_layout(struct inode *inode)
1049 {
1050 struct pnfs_layout_hdr *lo;
1051 int ret;
1052
1053 spin_lock(&inode->i_lock);
1054 lo = NFS_I(inode)->layout;
1055 if (lo == NULL) {
1056 spin_unlock(&inode->i_lock);
1057 return 0;
1058 }
1059 pnfs_get_layout_hdr(lo);
1060 /* Block new layoutgets and read/write to ds */
1061 lo->plh_block_lgets++;
1062 spin_unlock(&inode->i_lock);
1063 filemap_fdatawait(inode->i_mapping);
1064 ret = pnfs_layoutcommit_inode(inode, true);
1065 if (ret == 0)
1066 ret = _pnfs_return_layout(inode);
1067 spin_lock(&inode->i_lock);
1068 lo->plh_block_lgets--;
1069 spin_unlock(&inode->i_lock);
1070 pnfs_put_layout_hdr(lo);
1071 return ret;
1072 }
1073
1074 bool pnfs_roc(struct inode *ino)
1075 {
1076 struct nfs_inode *nfsi = NFS_I(ino);
1077 struct nfs_open_context *ctx;
1078 struct nfs4_state *state;
1079 struct pnfs_layout_hdr *lo;
1080 struct pnfs_layout_segment *lseg, *tmp;
1081 nfs4_stateid stateid;
1082 LIST_HEAD(tmp_list);
1083 bool found = false, layoutreturn = false;
1084
1085 spin_lock(&ino->i_lock);
1086 lo = nfsi->layout;
1087 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
1088 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1089 goto out_noroc;
1090
1091 /* Don't return layout if we hold a delegation */
1092 if (nfs4_check_delegation(ino, FMODE_READ))
1093 goto out_noroc;
1094
1095 list_for_each_entry(ctx, &nfsi->open_files, list) {
1096 state = ctx->state;
1097 /* Don't return layout if there is open file state */
1098 if (state != NULL && state->state != 0)
1099 goto out_noroc;
1100 }
1101
1102 pnfs_clear_retry_layoutget(lo);
1103 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1104 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
1105 mark_lseg_invalid(lseg, &tmp_list);
1106 found = true;
1107 }
1108 if (!found)
1109 goto out_noroc;
1110 lo->plh_block_lgets++;
1111 pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
1112 spin_unlock(&ino->i_lock);
1113 pnfs_free_lseg_list(&tmp_list);
1114 pnfs_layoutcommit_inode(ino, true);
1115 return true;
1116
1117 out_noroc:
1118 if (lo) {
1119 stateid = lo->plh_stateid;
1120 if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1121 &lo->plh_flags))
1122 layoutreturn = pnfs_prepare_layoutreturn(lo);
1123 }
1124 spin_unlock(&ino->i_lock);
1125 if (layoutreturn) {
1126 pnfs_layoutcommit_inode(ino, true);
1127 pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
1128 }
1129 return false;
1130 }
1131
1132 void pnfs_roc_release(struct inode *ino)
1133 {
1134 struct pnfs_layout_hdr *lo;
1135
1136 spin_lock(&ino->i_lock);
1137 lo = NFS_I(ino)->layout;
1138 lo->plh_block_lgets--;
1139 if (atomic_dec_and_test(&lo->plh_refcount)) {
1140 pnfs_detach_layout_hdr(lo);
1141 spin_unlock(&ino->i_lock);
1142 pnfs_free_layout_hdr(lo);
1143 } else
1144 spin_unlock(&ino->i_lock);
1145 }
1146
1147 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
1148 {
1149 struct pnfs_layout_hdr *lo;
1150
1151 spin_lock(&ino->i_lock);
1152 lo = NFS_I(ino)->layout;
1153 if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
1154 lo->plh_barrier = barrier;
1155 spin_unlock(&ino->i_lock);
1156 }
1157
1158 bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
1159 {
1160 struct nfs_inode *nfsi = NFS_I(ino);
1161 struct pnfs_layout_hdr *lo;
1162 struct pnfs_layout_segment *lseg;
1163 nfs4_stateid stateid;
1164 u32 current_seqid;
1165 bool layoutreturn = false;
1166
1167 spin_lock(&ino->i_lock);
1168 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) {
1169 if (!test_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1170 continue;
1171 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
1172 continue;
1173 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1174 spin_unlock(&ino->i_lock);
1175 return true;
1176 }
1177 lo = nfsi->layout;
1178 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
1179
1180 /* Since close does not return a layout stateid for use as
1181 * a barrier, we choose the worst-case barrier.
1182 */
1183 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1184 stateid = lo->plh_stateid;
1185 if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1186 &lo->plh_flags))
1187 layoutreturn = pnfs_prepare_layoutreturn(lo);
1188 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
1189 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1190
1191 spin_unlock(&ino->i_lock);
1192 if (layoutreturn) {
1193 pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, false);
1194 return true;
1195 }
1196 return false;
1197 }
1198
1199 /*
1200 * Compare two layout segments for sorting into layout cache.
1201 * We want to preferentially return RW over RO layouts, so ensure those
1202 * are seen first.
1203 */
1204 static s64
1205 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1206 const struct pnfs_layout_range *l2)
1207 {
1208 s64 d;
1209
1210 /* high offset > low offset */
1211 d = l1->offset - l2->offset;
1212 if (d)
1213 return d;
1214
1215 /* short length > long length */
1216 d = l2->length - l1->length;
1217 if (d)
1218 return d;
1219
1220 /* read > read/write */
1221 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1222 }
1223
1224 static void
1225 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1226 struct pnfs_layout_segment *lseg)
1227 {
1228 struct pnfs_layout_segment *lp;
1229
1230 dprintk("%s:Begin\n", __func__);
1231
1232 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
1233 if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
1234 continue;
1235 list_add_tail(&lseg->pls_list, &lp->pls_list);
1236 dprintk("%s: inserted lseg %p "
1237 "iomode %d offset %llu length %llu before "
1238 "lp %p iomode %d offset %llu length %llu\n",
1239 __func__, lseg, lseg->pls_range.iomode,
1240 lseg->pls_range.offset, lseg->pls_range.length,
1241 lp, lp->pls_range.iomode, lp->pls_range.offset,
1242 lp->pls_range.length);
1243 goto out;
1244 }
1245 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1246 dprintk("%s: inserted lseg %p "
1247 "iomode %d offset %llu length %llu at tail\n",
1248 __func__, lseg, lseg->pls_range.iomode,
1249 lseg->pls_range.offset, lseg->pls_range.length);
1250 out:
1251 pnfs_get_layout_hdr(lo);
1252
1253 dprintk("%s:Return\n", __func__);
1254 }
1255
1256 static struct pnfs_layout_hdr *
1257 alloc_init_layout_hdr(struct inode *ino,
1258 struct nfs_open_context *ctx,
1259 gfp_t gfp_flags)
1260 {
1261 struct pnfs_layout_hdr *lo;
1262
1263 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1264 if (!lo)
1265 return NULL;
1266 atomic_set(&lo->plh_refcount, 1);
1267 INIT_LIST_HEAD(&lo->plh_layouts);
1268 INIT_LIST_HEAD(&lo->plh_segs);
1269 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1270 lo->plh_inode = ino;
1271 lo->plh_lc_cred = get_rpccred(ctx->cred);
1272 return lo;
1273 }
1274
1275 static struct pnfs_layout_hdr *
1276 pnfs_find_alloc_layout(struct inode *ino,
1277 struct nfs_open_context *ctx,
1278 gfp_t gfp_flags)
1279 {
1280 struct nfs_inode *nfsi = NFS_I(ino);
1281 struct pnfs_layout_hdr *new = NULL;
1282
1283 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1284
1285 if (nfsi->layout != NULL)
1286 goto out_existing;
1287 spin_unlock(&ino->i_lock);
1288 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1289 spin_lock(&ino->i_lock);
1290
1291 if (likely(nfsi->layout == NULL)) { /* Won the race? */
1292 nfsi->layout = new;
1293 return new;
1294 } else if (new != NULL)
1295 pnfs_free_layout_hdr(new);
1296 out_existing:
1297 pnfs_get_layout_hdr(nfsi->layout);
1298 return nfsi->layout;
1299 }
1300
1301 /*
1302 * iomode matching rules:
1303 * iomode lseg match
1304 * ----- ----- -----
1305 * ANY READ true
1306 * ANY RW true
1307 * RW READ false
1308 * RW RW true
1309 * READ READ true
1310 * READ RW true
1311 */
1312 static bool
1313 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1314 const struct pnfs_layout_range *range)
1315 {
1316 struct pnfs_layout_range range1;
1317
1318 if ((range->iomode == IOMODE_RW &&
1319 ls_range->iomode != IOMODE_RW) ||
1320 !pnfs_lseg_range_intersecting(ls_range, range))
1321 return 0;
1322
1323 /* range1 covers only the first byte in the range */
1324 range1 = *range;
1325 range1.length = 1;
1326 return pnfs_lseg_range_contained(ls_range, &range1);
1327 }
1328
1329 /*
1330 * lookup range in layout
1331 */
1332 static struct pnfs_layout_segment *
1333 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1334 struct pnfs_layout_range *range)
1335 {
1336 struct pnfs_layout_segment *lseg, *ret = NULL;
1337
1338 dprintk("%s:Begin\n", __func__);
1339
1340 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1341 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1342 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1343 pnfs_lseg_range_match(&lseg->pls_range, range)) {
1344 ret = pnfs_get_lseg(lseg);
1345 break;
1346 }
1347 if (lseg->pls_range.offset > range->offset)
1348 break;
1349 }
1350
1351 dprintk("%s:Return lseg %p ref %d\n",
1352 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1353 return ret;
1354 }
1355
1356 /*
1357 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1358 * to the MDS or over pNFS
1359 *
1360 * The nfs_inode read_io and write_io fields are cumulative counters reset
1361 * when there are no layout segments. Note that in pnfs_update_layout iomode
1362 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1363 * WRITE request.
1364 *
1365 * A return of true means use MDS I/O.
1366 *
1367 * From rfc 5661:
1368 * If a file's size is smaller than the file size threshold, data accesses
1369 * SHOULD be sent to the metadata server. If an I/O request has a length that
1370 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1371 * server. If both file size and I/O size are provided, the client SHOULD
1372 * reach or exceed both thresholds before sending its read or write
1373 * requests to the data server.
1374 */
1375 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1376 struct inode *ino, int iomode)
1377 {
1378 struct nfs4_threshold *t = ctx->mdsthreshold;
1379 struct nfs_inode *nfsi = NFS_I(ino);
1380 loff_t fsize = i_size_read(ino);
1381 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1382
1383 if (t == NULL)
1384 return ret;
1385
1386 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1387 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1388
1389 switch (iomode) {
1390 case IOMODE_READ:
1391 if (t->bm & THRESHOLD_RD) {
1392 dprintk("%s fsize %llu\n", __func__, fsize);
1393 size_set = true;
1394 if (fsize < t->rd_sz)
1395 size = true;
1396 }
1397 if (t->bm & THRESHOLD_RD_IO) {
1398 dprintk("%s nfsi->read_io %llu\n", __func__,
1399 nfsi->read_io);
1400 io_set = true;
1401 if (nfsi->read_io < t->rd_io_sz)
1402 io = true;
1403 }
1404 break;
1405 case IOMODE_RW:
1406 if (t->bm & THRESHOLD_WR) {
1407 dprintk("%s fsize %llu\n", __func__, fsize);
1408 size_set = true;
1409 if (fsize < t->wr_sz)
1410 size = true;
1411 }
1412 if (t->bm & THRESHOLD_WR_IO) {
1413 dprintk("%s nfsi->write_io %llu\n", __func__,
1414 nfsi->write_io);
1415 io_set = true;
1416 if (nfsi->write_io < t->wr_io_sz)
1417 io = true;
1418 }
1419 break;
1420 }
1421 if (size_set && io_set) {
1422 if (size && io)
1423 ret = true;
1424 } else if (size || io)
1425 ret = true;
1426
1427 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1428 return ret;
1429 }
1430
1431 /* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */
1432 static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key)
1433 {
1434 if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags))
1435 return 1;
1436 return nfs_wait_bit_killable(key);
1437 }
1438
1439 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1440 {
1441 /*
1442 * send layoutcommit as it can hold up layoutreturn due to lseg
1443 * reference
1444 */
1445 pnfs_layoutcommit_inode(lo->plh_inode, false);
1446 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1447 pnfs_layoutget_retry_bit_wait,
1448 TASK_UNINTERRUPTIBLE);
1449 }
1450
1451 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1452 {
1453 unsigned long *bitlock = &lo->plh_flags;
1454
1455 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1456 smp_mb__after_atomic();
1457 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1458 }
1459
1460 /*
1461 * Layout segment is retreived from the server if not cached.
1462 * The appropriate layout segment is referenced and returned to the caller.
1463 */
1464 struct pnfs_layout_segment *
1465 pnfs_update_layout(struct inode *ino,
1466 struct nfs_open_context *ctx,
1467 loff_t pos,
1468 u64 count,
1469 enum pnfs_iomode iomode,
1470 gfp_t gfp_flags)
1471 {
1472 struct pnfs_layout_range arg = {
1473 .iomode = iomode,
1474 .offset = pos,
1475 .length = count,
1476 };
1477 unsigned pg_offset;
1478 struct nfs_server *server = NFS_SERVER(ino);
1479 struct nfs_client *clp = server->nfs_client;
1480 struct pnfs_layout_hdr *lo;
1481 struct pnfs_layout_segment *lseg = NULL;
1482 bool first;
1483
1484 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1485 goto out;
1486
1487 if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1488 goto out;
1489
1490 lookup_again:
1491 first = false;
1492 spin_lock(&ino->i_lock);
1493 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1494 if (lo == NULL) {
1495 spin_unlock(&ino->i_lock);
1496 goto out;
1497 }
1498
1499 /* Do we even need to bother with this? */
1500 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1501 dprintk("%s matches recall, use MDS\n", __func__);
1502 goto out_unlock;
1503 }
1504
1505 /* if LAYOUTGET already failed once we don't try again */
1506 if (pnfs_layout_io_test_failed(lo, iomode) &&
1507 !pnfs_should_retry_layoutget(lo))
1508 goto out_unlock;
1509
1510 first = list_empty(&lo->plh_segs);
1511 if (first) {
1512 /* The first layoutget for the file. Need to serialize per
1513 * RFC 5661 Errata 3208.
1514 */
1515 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1516 &lo->plh_flags)) {
1517 spin_unlock(&ino->i_lock);
1518 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1519 TASK_UNINTERRUPTIBLE);
1520 pnfs_put_layout_hdr(lo);
1521 goto lookup_again;
1522 }
1523 } else {
1524 /* Check to see if the layout for the given range
1525 * already exists
1526 */
1527 lseg = pnfs_find_lseg(lo, &arg);
1528 if (lseg)
1529 goto out_unlock;
1530 }
1531
1532 /*
1533 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1534 * for LAYOUTRETURN even if first is true.
1535 */
1536 if (!lseg && pnfs_should_retry_layoutget(lo) &&
1537 test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1538 spin_unlock(&ino->i_lock);
1539 dprintk("%s wait for layoutreturn\n", __func__);
1540 if (pnfs_prepare_to_retry_layoutget(lo)) {
1541 if (first)
1542 pnfs_clear_first_layoutget(lo);
1543 pnfs_put_layout_hdr(lo);
1544 dprintk("%s retrying\n", __func__);
1545 goto lookup_again;
1546 }
1547 goto out_put_layout_hdr;
1548 }
1549
1550 if (pnfs_layoutgets_blocked(lo, &arg, 0))
1551 goto out_unlock;
1552 atomic_inc(&lo->plh_outstanding);
1553 spin_unlock(&ino->i_lock);
1554
1555 if (list_empty(&lo->plh_layouts)) {
1556 /* The lo must be on the clp list if there is any
1557 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1558 */
1559 spin_lock(&clp->cl_lock);
1560 if (list_empty(&lo->plh_layouts))
1561 list_add_tail(&lo->plh_layouts, &server->layouts);
1562 spin_unlock(&clp->cl_lock);
1563 }
1564
1565 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1566 if (pg_offset) {
1567 arg.offset -= pg_offset;
1568 arg.length += pg_offset;
1569 }
1570 if (arg.length != NFS4_MAX_UINT64)
1571 arg.length = PAGE_CACHE_ALIGN(arg.length);
1572
1573 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1574 pnfs_clear_retry_layoutget(lo);
1575 atomic_dec(&lo->plh_outstanding);
1576 out_put_layout_hdr:
1577 if (first)
1578 pnfs_clear_first_layoutget(lo);
1579 pnfs_put_layout_hdr(lo);
1580 out:
1581 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1582 "(%s, offset: %llu, length: %llu)\n",
1583 __func__, ino->i_sb->s_id,
1584 (unsigned long long)NFS_FILEID(ino),
1585 lseg == NULL ? "not found" : "found",
1586 iomode==IOMODE_RW ? "read/write" : "read-only",
1587 (unsigned long long)pos,
1588 (unsigned long long)count);
1589 return lseg;
1590 out_unlock:
1591 spin_unlock(&ino->i_lock);
1592 goto out_put_layout_hdr;
1593 }
1594 EXPORT_SYMBOL_GPL(pnfs_update_layout);
1595
1596 struct pnfs_layout_segment *
1597 pnfs_layout_process(struct nfs4_layoutget *lgp)
1598 {
1599 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1600 struct nfs4_layoutget_res *res = &lgp->res;
1601 struct pnfs_layout_segment *lseg;
1602 struct inode *ino = lo->plh_inode;
1603 LIST_HEAD(free_me);
1604 int status = 0;
1605
1606 /* Inject layout blob into I/O device driver */
1607 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1608 if (!lseg || IS_ERR(lseg)) {
1609 if (!lseg)
1610 status = -ENOMEM;
1611 else
1612 status = PTR_ERR(lseg);
1613 dprintk("%s: Could not allocate layout: error %d\n",
1614 __func__, status);
1615 goto out;
1616 }
1617
1618 init_lseg(lo, lseg);
1619 lseg->pls_range = res->range;
1620
1621 spin_lock(&ino->i_lock);
1622 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1623 dprintk("%s forget reply due to recall\n", __func__);
1624 goto out_forget_reply;
1625 }
1626
1627 if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) {
1628 dprintk("%s forget reply due to state\n", __func__);
1629 goto out_forget_reply;
1630 }
1631
1632 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
1633 /* existing state ID, make sure the sequence number matches. */
1634 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1635 dprintk("%s forget reply due to sequence\n", __func__);
1636 goto out_forget_reply;
1637 }
1638 pnfs_set_layout_stateid(lo, &res->stateid, false);
1639 } else {
1640 /*
1641 * We got an entirely new state ID. Mark all segments for the
1642 * inode invalid, and don't bother validating the stateid
1643 * sequence number.
1644 */
1645 pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
1646
1647 nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
1648 lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
1649 }
1650
1651 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1652
1653 pnfs_get_lseg(lseg);
1654 pnfs_layout_insert_lseg(lo, lseg);
1655
1656 if (res->return_on_close) {
1657 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1658 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1659 }
1660
1661 spin_unlock(&ino->i_lock);
1662 pnfs_free_lseg_list(&free_me);
1663 return lseg;
1664 out:
1665 return ERR_PTR(status);
1666
1667 out_forget_reply:
1668 spin_unlock(&ino->i_lock);
1669 lseg->pls_layout = lo;
1670 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1671 goto out;
1672 }
1673
1674 static void
1675 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1676 struct list_head *tmp_list,
1677 struct pnfs_layout_range *return_range)
1678 {
1679 struct pnfs_layout_segment *lseg, *next;
1680
1681 dprintk("%s:Begin lo %p\n", __func__, lo);
1682
1683 if (list_empty(&lo->plh_segs))
1684 return;
1685
1686 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
1687 if (should_free_lseg(&lseg->pls_range, return_range)) {
1688 dprintk("%s: marking lseg %p iomode %d "
1689 "offset %llu length %llu\n", __func__,
1690 lseg, lseg->pls_range.iomode,
1691 lseg->pls_range.offset,
1692 lseg->pls_range.length);
1693 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1694 mark_lseg_invalid(lseg, tmp_list);
1695 }
1696 }
1697
1698 void pnfs_error_mark_layout_for_return(struct inode *inode,
1699 struct pnfs_layout_segment *lseg)
1700 {
1701 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
1702 int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode);
1703 struct pnfs_layout_range range = {
1704 .iomode = lseg->pls_range.iomode,
1705 .offset = 0,
1706 .length = NFS4_MAX_UINT64,
1707 };
1708 LIST_HEAD(free_me);
1709
1710 spin_lock(&inode->i_lock);
1711 /* set failure bit so that pnfs path will be retried later */
1712 pnfs_layout_set_fail_bit(lo, iomode);
1713 if (lo->plh_return_iomode == 0)
1714 lo->plh_return_iomode = range.iomode;
1715 else if (lo->plh_return_iomode != range.iomode)
1716 lo->plh_return_iomode = IOMODE_ANY;
1717 /*
1718 * mark all matching lsegs so that we are sure to have no live
1719 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1720 * for how it works.
1721 */
1722 pnfs_mark_matching_lsegs_return(lo, &free_me, &range);
1723 spin_unlock(&inode->i_lock);
1724 pnfs_free_lseg_list(&free_me);
1725 }
1726 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
1727
1728 void
1729 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1730 {
1731 u64 rd_size = req->wb_bytes;
1732
1733 if (pgio->pg_lseg == NULL) {
1734 if (pgio->pg_dreq == NULL)
1735 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1736 else
1737 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1738
1739 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1740 req->wb_context,
1741 req_offset(req),
1742 rd_size,
1743 IOMODE_READ,
1744 GFP_KERNEL);
1745 }
1746 /* If no lseg, fall back to read through mds */
1747 if (pgio->pg_lseg == NULL)
1748 nfs_pageio_reset_read_mds(pgio);
1749
1750 }
1751 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1752
1753 void
1754 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1755 struct nfs_page *req, u64 wb_size)
1756 {
1757 if (pgio->pg_lseg == NULL)
1758 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1759 req->wb_context,
1760 req_offset(req),
1761 wb_size,
1762 IOMODE_RW,
1763 GFP_NOFS);
1764 /* If no lseg, fall back to write through mds */
1765 if (pgio->pg_lseg == NULL)
1766 nfs_pageio_reset_write_mds(pgio);
1767 }
1768 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1769
1770 void
1771 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
1772 {
1773 if (desc->pg_lseg) {
1774 pnfs_put_lseg(desc->pg_lseg);
1775 desc->pg_lseg = NULL;
1776 }
1777 }
1778 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
1779
1780 /*
1781 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1782 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1783 */
1784 size_t
1785 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
1786 struct nfs_page *prev, struct nfs_page *req)
1787 {
1788 unsigned int size;
1789 u64 seg_end, req_start, seg_left;
1790
1791 size = nfs_generic_pg_test(pgio, prev, req);
1792 if (!size)
1793 return 0;
1794
1795 /*
1796 * 'size' contains the number of bytes left in the current page (up
1797 * to the original size asked for in @req->wb_bytes).
1798 *
1799 * Calculate how many bytes are left in the layout segment
1800 * and if there are less bytes than 'size', return that instead.
1801 *
1802 * Please also note that 'end_offset' is actually the offset of the
1803 * first byte that lies outside the pnfs_layout_range. FIXME?
1804 *
1805 */
1806 if (pgio->pg_lseg) {
1807 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1808 pgio->pg_lseg->pls_range.length);
1809 req_start = req_offset(req);
1810 WARN_ON_ONCE(req_start >= seg_end);
1811 /* start of request is past the last byte of this segment */
1812 if (req_start >= seg_end) {
1813 /* reference the new lseg */
1814 if (pgio->pg_ops->pg_cleanup)
1815 pgio->pg_ops->pg_cleanup(pgio);
1816 if (pgio->pg_ops->pg_init)
1817 pgio->pg_ops->pg_init(pgio, req);
1818 return 0;
1819 }
1820
1821 /* adjust 'size' iff there are fewer bytes left in the
1822 * segment than what nfs_generic_pg_test returned */
1823 seg_left = seg_end - req_start;
1824 if (seg_left < size)
1825 size = (unsigned int)seg_left;
1826 }
1827
1828 return size;
1829 }
1830 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1831
1832 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1833 {
1834 struct nfs_pageio_descriptor pgio;
1835
1836 /* Resend all requests through the MDS */
1837 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
1838 hdr->completion_ops);
1839 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
1840 return nfs_pageio_resend(&pgio, hdr);
1841 }
1842 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1843
1844 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1845 {
1846
1847 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1848 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1849 PNFS_LAYOUTRET_ON_ERROR) {
1850 pnfs_return_layout(hdr->inode);
1851 }
1852 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1853 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1854 }
1855
1856 /*
1857 * Called by non rpc-based layout drivers
1858 */
1859 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1860 {
1861 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1862 if (!hdr->pnfs_error) {
1863 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1864 hdr->mds_offset + hdr->res.count);
1865 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1866 } else
1867 pnfs_ld_handle_write_error(hdr);
1868 hdr->mds_ops->rpc_release(hdr);
1869 }
1870 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1871
1872 static void
1873 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1874 struct nfs_pgio_header *hdr)
1875 {
1876 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1877
1878 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1879 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1880 nfs_pageio_reset_write_mds(desc);
1881 mirror->pg_recoalesce = 1;
1882 }
1883 nfs_pgio_data_destroy(hdr);
1884 hdr->release(hdr);
1885 }
1886
1887 static enum pnfs_try_status
1888 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1889 const struct rpc_call_ops *call_ops,
1890 struct pnfs_layout_segment *lseg,
1891 int how)
1892 {
1893 struct inode *inode = hdr->inode;
1894 enum pnfs_try_status trypnfs;
1895 struct nfs_server *nfss = NFS_SERVER(inode);
1896
1897 hdr->mds_ops = call_ops;
1898
1899 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1900 inode->i_ino, hdr->args.count, hdr->args.offset, how);
1901 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1902 if (trypnfs != PNFS_NOT_ATTEMPTED)
1903 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1904 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1905 return trypnfs;
1906 }
1907
1908 static void
1909 pnfs_do_write(struct nfs_pageio_descriptor *desc,
1910 struct nfs_pgio_header *hdr, int how)
1911 {
1912 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1913 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1914 enum pnfs_try_status trypnfs;
1915
1916 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
1917 if (trypnfs == PNFS_NOT_ATTEMPTED)
1918 pnfs_write_through_mds(desc, hdr);
1919 }
1920
1921 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1922 {
1923 pnfs_put_lseg(hdr->lseg);
1924 nfs_pgio_header_free(hdr);
1925 }
1926
1927 int
1928 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1929 {
1930 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1931
1932 struct nfs_pgio_header *hdr;
1933 int ret;
1934
1935 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
1936 if (!hdr) {
1937 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
1938 return -ENOMEM;
1939 }
1940 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1941
1942 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1943 ret = nfs_generic_pgio(desc, hdr);
1944 if (!ret)
1945 pnfs_do_write(desc, hdr, desc->pg_ioflags);
1946
1947 return ret;
1948 }
1949 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1950
1951 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
1952 {
1953 struct nfs_pageio_descriptor pgio;
1954
1955 /* Resend all requests through the MDS */
1956 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
1957 return nfs_pageio_resend(&pgio, hdr);
1958 }
1959 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1960
1961 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1962 {
1963 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1964 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1965 PNFS_LAYOUTRET_ON_ERROR) {
1966 pnfs_return_layout(hdr->inode);
1967 }
1968 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1969 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
1970 }
1971
1972 /*
1973 * Called by non rpc-based layout drivers
1974 */
1975 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
1976 {
1977 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
1978 if (likely(!hdr->pnfs_error)) {
1979 __nfs4_read_done_cb(hdr);
1980 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1981 } else
1982 pnfs_ld_handle_read_error(hdr);
1983 hdr->mds_ops->rpc_release(hdr);
1984 }
1985 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1986
1987 static void
1988 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1989 struct nfs_pgio_header *hdr)
1990 {
1991 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1992
1993 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1994 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1995 nfs_pageio_reset_read_mds(desc);
1996 mirror->pg_recoalesce = 1;
1997 }
1998 nfs_pgio_data_destroy(hdr);
1999 hdr->release(hdr);
2000 }
2001
2002 /*
2003 * Call the appropriate parallel I/O subsystem read function.
2004 */
2005 static enum pnfs_try_status
2006 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2007 const struct rpc_call_ops *call_ops,
2008 struct pnfs_layout_segment *lseg)
2009 {
2010 struct inode *inode = hdr->inode;
2011 struct nfs_server *nfss = NFS_SERVER(inode);
2012 enum pnfs_try_status trypnfs;
2013
2014 hdr->mds_ops = call_ops;
2015
2016 dprintk("%s: Reading ino:%lu %u@%llu\n",
2017 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2018
2019 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2020 if (trypnfs != PNFS_NOT_ATTEMPTED)
2021 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2022 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2023 return trypnfs;
2024 }
2025
2026 /* Resend all requests through pnfs. */
2027 int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2028 {
2029 struct nfs_pageio_descriptor pgio;
2030
2031 nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops);
2032 return nfs_pageio_resend(&pgio, hdr);
2033 }
2034 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2035
2036 static void
2037 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2038 {
2039 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2040 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2041 enum pnfs_try_status trypnfs;
2042 int err = 0;
2043
2044 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2045 if (trypnfs == PNFS_TRY_AGAIN)
2046 err = pnfs_read_resend_pnfs(hdr);
2047 if (trypnfs == PNFS_NOT_ATTEMPTED || err)
2048 pnfs_read_through_mds(desc, hdr);
2049 }
2050
2051 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2052 {
2053 pnfs_put_lseg(hdr->lseg);
2054 nfs_pgio_header_free(hdr);
2055 }
2056
2057 int
2058 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2059 {
2060 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2061
2062 struct nfs_pgio_header *hdr;
2063 int ret;
2064
2065 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2066 if (!hdr) {
2067 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
2068 return -ENOMEM;
2069 }
2070 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2071 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2072 ret = nfs_generic_pgio(desc, hdr);
2073 if (!ret)
2074 pnfs_do_read(desc, hdr);
2075 return ret;
2076 }
2077 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2078
2079 static void pnfs_clear_layoutcommitting(struct inode *inode)
2080 {
2081 unsigned long *bitlock = &NFS_I(inode)->flags;
2082
2083 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2084 smp_mb__after_atomic();
2085 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2086 }
2087
2088 /*
2089 * There can be multiple RW segments.
2090 */
2091 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2092 {
2093 struct pnfs_layout_segment *lseg;
2094
2095 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2096 if (lseg->pls_range.iomode == IOMODE_RW &&
2097 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2098 list_add(&lseg->pls_lc_list, listp);
2099 }
2100 }
2101
2102 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2103 {
2104 struct pnfs_layout_segment *lseg, *tmp;
2105
2106 /* Matched by references in pnfs_set_layoutcommit */
2107 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2108 list_del_init(&lseg->pls_lc_list);
2109 pnfs_put_lseg(lseg);
2110 }
2111
2112 pnfs_clear_layoutcommitting(inode);
2113 }
2114
2115 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2116 {
2117 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2118 }
2119 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2120
2121 void
2122 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2123 loff_t end_pos)
2124 {
2125 struct nfs_inode *nfsi = NFS_I(inode);
2126 bool mark_as_dirty = false;
2127
2128 spin_lock(&inode->i_lock);
2129 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2130 nfsi->layout->plh_lwb = end_pos;
2131 mark_as_dirty = true;
2132 dprintk("%s: Set layoutcommit for inode %lu ",
2133 __func__, inode->i_ino);
2134 } else if (end_pos > nfsi->layout->plh_lwb)
2135 nfsi->layout->plh_lwb = end_pos;
2136 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2137 /* references matched in nfs4_layoutcommit_release */
2138 pnfs_get_lseg(lseg);
2139 }
2140 spin_unlock(&inode->i_lock);
2141 dprintk("%s: lseg %p end_pos %llu\n",
2142 __func__, lseg, nfsi->layout->plh_lwb);
2143
2144 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2145 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2146 if (mark_as_dirty)
2147 mark_inode_dirty_sync(inode);
2148 }
2149 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2150
2151 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2152 {
2153 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2154
2155 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2156 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2157 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2158 }
2159
2160 /*
2161 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2162 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2163 * data to disk to allow the server to recover the data if it crashes.
2164 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2165 * is off, and a COMMIT is sent to a data server, or
2166 * if WRITEs to a data server return NFS_DATA_SYNC.
2167 */
2168 int
2169 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2170 {
2171 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2172 struct nfs4_layoutcommit_data *data;
2173 struct nfs_inode *nfsi = NFS_I(inode);
2174 loff_t end_pos;
2175 int status;
2176
2177 if (!pnfs_layoutcommit_outstanding(inode))
2178 return 0;
2179
2180 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2181
2182 status = -EAGAIN;
2183 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2184 if (!sync)
2185 goto out;
2186 status = wait_on_bit_lock_action(&nfsi->flags,
2187 NFS_INO_LAYOUTCOMMITTING,
2188 nfs_wait_bit_killable,
2189 TASK_KILLABLE);
2190 if (status)
2191 goto out;
2192 }
2193
2194 status = -ENOMEM;
2195 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2196 data = kzalloc(sizeof(*data), GFP_NOFS);
2197 if (!data)
2198 goto clear_layoutcommitting;
2199
2200 status = 0;
2201 spin_lock(&inode->i_lock);
2202 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2203 goto out_unlock;
2204
2205 INIT_LIST_HEAD(&data->lseg_list);
2206 pnfs_list_write_lseg(inode, &data->lseg_list);
2207
2208 end_pos = nfsi->layout->plh_lwb;
2209
2210 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2211 spin_unlock(&inode->i_lock);
2212
2213 data->args.inode = inode;
2214 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
2215 nfs_fattr_init(&data->fattr);
2216 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2217 data->res.fattr = &data->fattr;
2218 data->args.lastbytewritten = end_pos - 1;
2219 data->res.server = NFS_SERVER(inode);
2220
2221 if (ld->prepare_layoutcommit) {
2222 status = ld->prepare_layoutcommit(&data->args);
2223 if (status) {
2224 put_rpccred(data->cred);
2225 spin_lock(&inode->i_lock);
2226 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2227 if (end_pos > nfsi->layout->plh_lwb)
2228 nfsi->layout->plh_lwb = end_pos;
2229 goto out_unlock;
2230 }
2231 }
2232
2233
2234 status = nfs4_proc_layoutcommit(data, sync);
2235 out:
2236 if (status)
2237 mark_inode_dirty_sync(inode);
2238 dprintk("<-- %s status %d\n", __func__, status);
2239 return status;
2240 out_unlock:
2241 spin_unlock(&inode->i_lock);
2242 kfree(data);
2243 clear_layoutcommitting:
2244 pnfs_clear_layoutcommitting(inode);
2245 goto out;
2246 }
2247 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2248
2249 int
2250 pnfs_generic_sync(struct inode *inode, bool datasync)
2251 {
2252 return pnfs_layoutcommit_inode(inode, true);
2253 }
2254 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2255
2256 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2257 {
2258 struct nfs4_threshold *thp;
2259
2260 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2261 if (!thp) {
2262 dprintk("%s mdsthreshold allocation failed\n", __func__);
2263 return NULL;
2264 }
2265 return thp;
2266 }
2267
2268 #if IS_ENABLED(CONFIG_NFS_V4_2)
2269 int
2270 pnfs_report_layoutstat(struct inode *inode)
2271 {
2272 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2273 struct nfs_server *server = NFS_SERVER(inode);
2274 struct nfs_inode *nfsi = NFS_I(inode);
2275 struct nfs42_layoutstat_data *data;
2276 struct pnfs_layout_hdr *hdr;
2277 int status = 0;
2278
2279 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2280 goto out;
2281
2282 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2283 goto out;
2284
2285 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2286 goto out;
2287
2288 spin_lock(&inode->i_lock);
2289 if (!NFS_I(inode)->layout) {
2290 spin_unlock(&inode->i_lock);
2291 goto out;
2292 }
2293 hdr = NFS_I(inode)->layout;
2294 pnfs_get_layout_hdr(hdr);
2295 spin_unlock(&inode->i_lock);
2296
2297 data = kzalloc(sizeof(*data), GFP_KERNEL);
2298 if (!data) {
2299 status = -ENOMEM;
2300 goto out_put;
2301 }
2302
2303 data->args.fh = NFS_FH(inode);
2304 data->args.inode = inode;
2305 nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid);
2306 status = ld->prepare_layoutstats(&data->args);
2307 if (status)
2308 goto out_free;
2309
2310 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2311
2312 out:
2313 dprintk("%s returns %d\n", __func__, status);
2314 return status;
2315
2316 out_free:
2317 kfree(data);
2318 out_put:
2319 pnfs_put_layout_hdr(hdr);
2320 smp_mb__before_atomic();
2321 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2322 smp_mb__after_atomic();
2323 goto out;
2324 }
2325 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2326 #endif
Linux with added FPC-III support